Skip to content
Search ⌘K
Get started

Tiger CLI and Tiger MCP

Understand Tiger CLI and Tiger MCP, how they compare to the REST API, and where they fit in your development lifecycle

Tiger CLI and Tiger MCP give you, your scripts, and your AI agent a programmatic control layer for Tiger Cloud. Instead of clicking through Console, you provision, configure, query, and manage services from your terminal or directly from tools like Claude Code, Cursor, and Codex.

This page explains what each tool is, how they relate, and where they fit in your workflow. To install and start using them, see Get started with Tiger CLI and Integrate Tiger Cloud with your AI agent.

What each tool is

Section titled “What each tool is”
  • Tiger CLI: a single binary that manages Tiger Cloud resources, including services, read replicas, VPCs, and related infrastructure. It is scriptable and verifiable, which makes it a good fit for automation, CI/CD, and day-to-day operations. Tiger CLI calls Tiger REST API under the hood.
  • Tiger MCP: a Model Context Protocol server bundled inside the Tiger CLI binary. It gives your AI agent tools to manage services and run SQL, plus built-in skills (for example, schema design, hypertable setup, and migration planning) and access to Tiger Data documentation. Beyond running operations, it can design schemas, analyze your database, and recommend improvements in plain language.
  • Tiger REST API: the underlying HTTP API. Use it directly when you build your own integrations or need language-native access from an application.
One install, three surfaces

Tiger MCP ships inside Tiger CLI, so a single tiger install gives you the command line and the MCP server. Both authenticate the same way and respect the same permissions.

When to use which

Section titled “When to use which”
You want to...Use
Script, automate, or run repeatable operations (CI/CD, pipelines)Tiger CLI
Work in natural language, or get design, analysis, and optimization helpTiger MCP
Build a custom integration or app in your own languageTiger REST API

They are complementary, not the same

Section titled “They are complementary, not the same”

Tiger CLI and Tiger MCP share authentication and can perform many of the same operations, but they are not interchangeable:

  • Only Tiger MCP can reason about your database. Using its built-in skills and the documentation, it designs schemas, finds hypertable candidates, plans migrations, and recommends improvements in plain language. The CLI has no command for this.
  • Only Tiger CLI gives you deterministic, scriptable execution: interactive psql sessions, connection strings, and repeatable commands for automation and CI/CD. Tiger MCP runs one statement at a time and, as an AI agent, is non-deterministic.

Use Tiger MCP for exploration, design, and analysis, and Tiger CLI for automation and precise, repeatable operations. Many workflows combine them: let an agent explore and suggest changes with Tiger MCP, then apply and verify them with CLI.

Where they fit in your lifecycle

Section titled “Where they fit in your lifecycle”

Tiger CLI and Tiger MCP support each stage of working with Tiger Cloud:

  1. Get started and set up: install the tools, authenticate, and connect your AI agent.
  2. Build: create services, design hypertables, add continuous aggregates, and configure jobs and policies.
  3. Validate: ask your agent to review a schema against best practices, or fork a service to test a change safely.
  4. Operate day to day: run queries, manage credentials, resize, and inspect service state.
  5. Monitor and debug: pull logs and investigate performance without leaving your terminal or agent.

Use agents safely

Section titled “Use agents safely”

Because Tiger MCP can act on your database, control what an agent is allowed to do:

  • Enable read-only mode so an agent can explore and query without making changes. See Best practices for AI agents.
  • Point agents at a read replica or a fork for exploratory work, so production is never at risk.
  • Prefer having the database do the work: ask for a single answer computed in-database rather than exporting raw rows to the client.

Next steps

Section titled “Next steps”